Design of missile longitudinal controller based on deep deterministic policy gradient learning algorithm

For the longitudinal control problem of missiles under highly nonlinear conditions, traditional controllers designed by fixed points have good static characteristics, but in the face of highly nonlinear systems, the control parameters designed at a certain characteristic point will not be able to satisfy the entire control system. In this paper, a combination of reinforcement learning and deep neural network architecture is used to propose an intelligent adaptive controller design scheme based on a deep deterministic policy gradient algorithm. Firstly, the control problem is reconstructed in the framework of deep reinforcement learning. The environment is composed of a two-degree-of-freedom nonlinear model of missile longitudinal dynamics. Secondly, the design of the reward function not only considers the stability of the longitudinal dynamics, but also uses the ITAE index as the reward function. The penalty item is to comprehensively consider the balance of error and execution output. Finally, simulation results are shown to check the ability of the proposed data-driven method to achieve the specified closed-loop response in a fully model-free manner under a complete flight condition, and the method is verified by model parameter-pull Monte Carlo simulations. Feasibility and effectiveness.